Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method comprising: intercepting a voice communication from a caller to a callee; collecting multi-sensory inputs associated with the voice communication by capturing data via at least one sensor device, applying at least one learned classifier to the captured data, and aggregating one or more classification results from the at least one learned classifier as the multi-sensory inputs, wherein the captured data comprises at least one biometric measure of the callee, and the multi-sensory inputs are indicative of content of the voice communication, at least one action carried out by the callee in response to the content of the voice communication, and a stress level of the callee in response to the content of the voice communication; and determining an overall risk assessment metric for the voice communication based on the multi-sensory inputs and learned signatures, wherein the overall risk assessment metric indicates a likelihood the voice communication is a scam.
The invention relates to a system for detecting fraudulent voice communications, such as scam calls, by analyzing multi-sensory inputs and biometric data associated with the call. The method involves intercepting a voice communication between a caller and a callee and capturing data from at least one sensor device. The captured data includes biometric measures of the callee, such as physiological responses, and other inputs that reflect the content of the call, the callee's actions in response to the call, and their stress level. A learned classifier processes this data to generate classification results, which are aggregated to form the multi-sensory inputs. These inputs are then used to determine an overall risk assessment metric, which evaluates the likelihood that the call is a scam. The system leverages learned signatures—predefined patterns or models—to assess risk based on the callee's reactions and biometric responses. The goal is to enhance fraud detection by analyzing not just the call content but also the callee's physiological and behavioral responses to identify potential scams.
2. The method of claim 1 , wherein the voice communication is one of a phone call or a voice message.
This invention relates to voice communication systems, specifically methods for processing and analyzing voice communications such as phone calls or voice messages. The technology addresses the challenge of efficiently handling and extracting meaningful data from voice interactions in real-time or near-real-time environments. The method involves capturing voice data from a communication session, which may be a live phone call or a recorded voice message. The system then processes this voice data to identify and extract relevant information, such as keywords, speaker identification, or sentiment analysis. The extracted data is then used to generate insights, trigger automated actions, or update databases. The method may also include filtering or enhancing the voice data to improve accuracy before analysis. The invention aims to streamline voice communication workflows by automating the extraction of useful information, reducing manual effort, and enabling faster decision-making based on voice interactions. The system can be integrated into customer service platforms, call centers, or other applications where voice data analysis is valuable. The method ensures that voice communications are processed efficiently, regardless of whether they are live calls or pre-recorded messages, and provides actionable insights from the spoken content.
3. The method of claim 1 , wherein the learned signatures comprises at least one of: a signature of a known scam, and a conversation signature of a trusted person of the callee.
This invention relates to fraud detection in telecommunications, specifically identifying scam calls by analyzing conversation patterns. The problem addressed is the difficulty in detecting sophisticated scam calls that mimic legitimate interactions, bypassing traditional caller ID or blacklist-based systems. The method involves using machine learning to generate learned signatures from call data. These signatures include patterns associated with known scam calls, such as specific phrases, call durations, or interaction sequences. Additionally, the method learns conversation signatures from trusted contacts of the callee, establishing a baseline for normal communication behavior. By comparing incoming calls against these learned signatures, the system can flag suspicious activity, such as calls that deviate from trusted patterns or match known scam signatures. The system processes audio or metadata from calls to extract features, which are then analyzed to detect anomalies or matches with known fraudulent patterns. The learned signatures are continuously updated to adapt to new scam tactics and evolving communication behaviors. This approach enhances fraud detection by leveraging behavioral analytics rather than relying solely on caller identification.
4. The method of claim 1 , further comprising: determining whether an intervention is necessary based on the overall risk assessment metric, the multi-sensory inputs, and a set of rules, wherein each rule specifies an event, a condition to satisfy if the event occurs, and a set of actions to perform if the condition is satisfied; and in response to determining an intervention is necessary, performing at least one of the following: providing a recommendation to the callee, providing a warning to the callee, providing a suggestion to the callee, or enforcing an action.
This invention relates to a system for assessing risk during a communication session, such as a phone call, to determine whether an intervention is necessary. The system collects multi-sensory inputs from the callee, including biometric data, environmental data, and behavioral data, to generate an overall risk assessment metric. The risk assessment metric quantifies the likelihood of a harmful event occurring to the callee. The system then evaluates this metric alongside the multi-sensory inputs and a predefined set of rules. Each rule defines an event, a condition that must be met if the event occurs, and a corresponding set of actions. If the conditions of any rule are satisfied, the system determines that an intervention is necessary. In response, the system may provide recommendations, warnings, or suggestions to the callee or enforce an action to mitigate the risk. The system dynamically adapts its interventions based on real-time data, ensuring timely and appropriate responses to potential threats. This approach enhances safety by proactively identifying and addressing risks during communication sessions.
5. The method of claim 1 , wherein the captured data comprises a recording of the voice communication captured by the at least one sensor device configured to continuously listen to, and initiate a recording of, one or more surrounding noises and a conversation of the callee.
This invention relates to a system for capturing and analyzing voice communications, particularly in scenarios where continuous monitoring of ambient sounds and conversations is required. The system employs at least one sensor device equipped to continuously listen to and record surrounding noises and conversations involving a callee. The sensor device is designed to initiate recordings automatically, ensuring that all relevant audio data is captured without manual intervention. This approach addresses the challenge of reliably documenting voice communications in environments where traditional recording methods may be inconsistent or require active user participation. The recorded data may include both the callee's conversation and background noises, providing a comprehensive audio log for analysis. The system is particularly useful in applications such as call centers, security monitoring, or healthcare, where accurate and continuous audio documentation is critical. By automating the recording process, the invention ensures that no relevant audio information is missed, improving the reliability and completeness of the captured data. The sensor device's continuous listening capability distinguishes it from systems that rely on user-triggered recordings, offering a more robust solution for environments where real-time audio capture is essential.
6. The method of claim 5 , wherein collecting multi-sensory inputs associated with the voice communication further comprises: obtaining a transcript of a conversation or speech included in the recording by transcribing the conversation or speech using a speech-to-text process; and applying one or more text-based learned models of the at least one learned classifier to the transcript to obtain the one or more classification results, wherein the one or more classification results comprise one or more predicted classification classes and one or more corresponding confidence levels, and the overall risk assessment metric is based on the one or more predicted classification classes and the one or more corresponding confidence levels.
This invention relates to voice communication analysis systems that assess risk by processing multi-sensory inputs, including audio recordings. The system addresses the challenge of detecting harmful or high-risk content in voice communications by combining speech transcription with machine learning-based classification. The method involves transcribing recorded conversations or speech into text using speech-to-text technology. The transcribed text is then analyzed using text-based learned models, which classify the content into predefined categories (e.g., threat levels, emotional states, or compliance violations). Each classification result includes a predicted class and a confidence level, which contribute to an overall risk assessment metric. The system integrates these classifications to determine the likelihood of risk, enabling automated monitoring and intervention in high-risk scenarios. The approach enhances traditional audio analysis by incorporating natural language processing and machine learning to improve accuracy and contextual understanding of spoken content. This method is particularly useful in applications like call center monitoring, security screening, or compliance enforcement, where real-time or post-call analysis of voice interactions is critical.
7. The method of claim 6 , wherein the one or more predicted classification classes comprise at least one of the following: a predicted scam type, a predicted conversation signature of a trusted person of the callee, a predicted caller identity of the caller, a predicted call purpose, a predicted social engineering tactic, and a list of suspicious keywords.
This invention relates to enhancing caller identification and fraud detection in telecommunications by analyzing call metadata and audio content to predict various aspects of a call. The problem addressed is the inability of traditional caller ID systems to accurately identify scams, impersonation attempts, or malicious intent during phone calls. The method involves processing call metadata, such as caller and callee information, call duration, and time of day, along with audio features extracted from the call, such as speech patterns, tone, and keywords. Machine learning models analyze these inputs to predict multiple classification classes, including the type of scam being attempted, the caller's identity, the purpose of the call, and the use of social engineering tactics. Additionally, the system identifies conversation signatures of trusted contacts to detect impersonation and generates a list of suspicious keywords. By combining these predictions, the system provides real-time alerts or call blocking to protect users from fraudulent or malicious calls. The approach improves upon existing caller ID systems by leveraging advanced analytics to detect sophisticated scams and deception techniques.
8. The method of claim 5 , wherein collecting multi-sensory inputs associated with the voice communication further comprises: applying one or more sound-based learned models of the at least one learned classifier to the recording to obtain the one or more classification results, wherein the one or more classification results comprise one or more predicted classification classes and one or more corresponding confidence levels, and the overall risk assessment metric is based on the one or more predicted classification classes and the one or more corresponding confidence levels.
This invention relates to voice communication analysis systems that assess risk by processing multi-sensory inputs. The technology addresses the challenge of detecting potential risks, such as fraud or security threats, during voice interactions by analyzing audio recordings and other sensory data. The system uses learned classifiers, including sound-based models, to evaluate voice communications. These classifiers generate classification results, which include predicted risk categories (e.g., fraudulent, suspicious, or safe) and associated confidence levels. The system then combines these results to produce an overall risk assessment metric, quantifying the likelihood of a threat. The method ensures that multiple sensory inputs, such as audio features, speaker characteristics, and contextual data, are integrated to improve accuracy. By leveraging machine learning models, the system dynamically adapts to different voice patterns and communication scenarios, enhancing threat detection capabilities. The approach is particularly useful in applications like call centers, security monitoring, and fraud prevention, where real-time risk assessment is critical. The invention improves upon traditional methods by incorporating confidence-weighted classifications, reducing false positives and negatives in risk detection.
9. The method of claim 8 , wherein the one or more predicted classification classes comprise at least one of the following: a predicted caller identity of the caller, a predicted emotion of the callee, and a predicted emotion of an adversary.
This invention relates to a method for analyzing and classifying audio signals, particularly in the context of voice communications such as phone calls. The method addresses the challenge of extracting meaningful insights from audio data to improve communication outcomes, such as identifying caller identity, detecting emotions of participants, and assessing adversarial intent. The method involves processing an audio signal from a communication session to generate a feature vector representing the audio data. This feature vector is then analyzed using a trained machine learning model to predict one or more classification classes. These classes include the predicted identity of the caller, the emotional state of the callee (the person receiving the call), and the emotional state of an adversary (if present). The predictions are derived from patterns in the audio signal, such as speech characteristics, tone, and other acoustic features. The machine learning model is trained using a dataset of labeled audio samples, where each sample is associated with known caller identities, emotions, or adversarial behaviors. The model learns to map these features to the corresponding classifications, enabling real-time or near-real-time predictions during live communications. The method may also include refining the model over time by incorporating new data to improve accuracy. This approach enhances communication systems by providing automated insights into caller identity and emotional states, which can be used for security, customer service, or conflict resolution applications. The system can flag suspicious callers, detect distress in callers or recipients, and identify adversarial tactics in real time.
10. The method of claim 1 , wherein the at least one sensor device comprises at least one of the following: one or more wearable corporal sensor devices configured to capture the at least one biometric measure of the callee, and one or more other sensor devices deployed in an object of the callee to detect the at least one action carried out by the callee.
This invention relates to systems for monitoring and analyzing biometric and behavioral data of a callee during a communication session, such as a phone call or video call. The problem addressed is the lack of real-time insights into the callee's physiological and behavioral state, which can be critical for applications like emergency response, healthcare monitoring, or security verification. The method involves using at least one sensor device to capture biometric measures of the callee, such as heart rate, respiration, or movement, and to detect actions performed by the callee, such as gestures or interactions with objects. The sensor devices may include wearable corporal sensors, such as smartwatches or fitness trackers, which monitor biometric data directly from the callee's body. Additionally, sensor devices may be embedded in objects the callee interacts with, such as smartphones, vehicles, or household items, to detect actions like button presses, movements, or environmental changes. The collected data is analyzed to determine the callee's state, such as stress levels, physical condition, or intent, and may be used to trigger automated responses, such as alerting emergency services or adjusting communication settings. The system improves situational awareness and enables proactive interventions based on real-time physiological and behavioral cues.
11. The method of claim 1 , wherein the at least one action carried out by the callee comprises at least one of: the callee using a user device to complete a financial transaction, the callee looking up personal information, the callee providing the caller with remote access to the user device, and the callee removing an object to carry out a financial related activity.
This invention relates to methods for enabling a callee to perform specific actions in response to a call, particularly in scenarios where the callee may be unable to directly interact with a caller. The problem addressed is the need for secure and efficient ways to facilitate actions such as financial transactions, information retrieval, or device access without requiring direct communication between the caller and callee. The method involves a callee receiving a call and, in response, performing at least one predefined action using a user device. These actions may include completing a financial transaction, such as a payment or transfer, using the device. The callee may also look up personal information stored on the device, such as account details or contact records. Additionally, the callee can provide the caller with remote access to the user device, allowing the caller to control or interact with the device remotely. Another possible action is the callee removing an object, such as a physical item or digital asset, to carry out a financial-related activity, such as withdrawing cash or transferring ownership. The method ensures that these actions are performed securely and efficiently, leveraging the callee's device to execute tasks that would otherwise require direct interaction. This is particularly useful in scenarios where the callee is unable to verbally communicate or when immediate action is required. The system may include authentication steps to verify the caller's identity before granting access or executing transactions.
12. A system comprising: at least one processor; and a non-transitory processor-readable memory device storing instructions that when executed by the at least one processor causes the at least one processor to perform operations including: intercepting a voice communication from a caller to a callee; collecting multi-sensory inputs associated with the voice communication by capturing data via at least one sensor device, applying at least one learned classifier to the captured data, and aggregating one or more classification results from the at least one learned classifier as the multi-sensory inputs, wherein the captured data comprises at least one biometric measure of the callee, and the multi-sensory inputs are indicative of content of the voice communication, at least one action carried out by the callee in response to the content of the voice communication, and a stress level of the callee in response to the content of the voice communication; and determining an overall risk assessment metric for the voice communication based on the multi-sensory inputs and learned signatures, wherein the overall risk assessment metric indicates a likelihood the voice communication is a scam.
This system detects and assesses the risk of scam voice communications by analyzing multi-sensory inputs and biometric data. The system intercepts a voice call between a caller and a callee, then collects data from sensors to capture biometric measures of the callee, such as heart rate or stress indicators. The system applies machine learning classifiers to this data to determine the content of the call, the callee's actions in response, and their stress level. These inputs are aggregated to generate an overall risk assessment metric, which indicates the likelihood that the call is a scam. The system uses learned signatures—predefined patterns or models—to evaluate the inputs and determine risk. By combining biometric analysis with contextual data, the system provides a comprehensive evaluation of potential fraudulent activity during voice communications. This approach enhances fraud detection by leveraging real-time physiological and behavioral responses to identify suspicious calls.
13. The system of claim 12 , wherein the learned signatures comprises at least one of: a signature of a known scam, and a conversation signature of a trusted person of the callee.
This invention relates to a system for detecting and preventing scam calls by analyzing communication patterns. The system addresses the problem of fraudulent calls that deceive recipients by mimicking legitimate conversations or impersonating trusted contacts. The system includes a database of learned signatures, which are unique patterns or characteristics extracted from prior communications. These signatures may include known scam call patterns, such as specific phrases, call durations, or interaction sequences, as well as conversation signatures of trusted individuals, such as speech patterns, vocabulary, or call behavior. The system compares incoming calls against these signatures to identify potential scams or unauthorized impersonations. If a match is detected, the system can alert the recipient, block the call, or take other protective actions. The system may also update its database dynamically as new scam patterns or trusted signatures are identified. This approach enhances security by leveraging machine learning and behavioral analysis to distinguish between legitimate and fraudulent communications.
14. The system of claim 12 , the operations further comprising: determining whether an intervention is necessary based on the overall risk assessment metric, the multi-sensory inputs, and a set of rules, wherein each rule specifies an event, a condition to satisfy if the event occurs, and a set of actions to perform if the condition is satisfied; and in response to determining an intervention is necessary, performing at least one of the following: providing a recommendation to the callee, providing a warning to the callee, providing a suggestion to the callee, or enforcing an action.
A system monitors and assesses risk during communication sessions, such as phone calls, to determine if intervention is necessary. The system collects multi-sensory inputs, including audio, video, and biometric data, to analyze the communication environment and participant behavior. It generates an overall risk assessment metric based on these inputs, which quantifies the likelihood of adverse events like harassment, distress, or safety threats. The system applies a set of predefined rules to evaluate the risk. Each rule defines an event (e.g., detection of raised voices), a condition (e.g., sustained for a specified duration), and corresponding actions (e.g., issuing a warning). If the risk assessment metric or rule conditions indicate a need for intervention, the system responds by providing recommendations, warnings, or suggestions to the callee. In severe cases, it may enforce actions such as terminating the call or alerting authorities. The system dynamically adapts its responses based on real-time data and rule evaluations to mitigate risks effectively. This approach enhances safety and well-being during communication by proactively identifying and addressing potential threats.
15. The system of claim 12 , wherein the captured data comprises a recording of the voice communication captured by the at least one sensor device configured to continuously listen to, and initiate a recording of, one or more surrounding noises and a conversation of the callee.
This invention relates to a system for capturing and processing voice communications, particularly in scenarios where continuous monitoring of ambient sounds and conversations is required. The system includes at least one sensor device designed to continuously listen to and record surrounding noises and conversations involving a callee. The sensor device is configured to initiate recordings automatically, ensuring that all relevant audio data is captured without manual intervention. The captured data includes both the voice communication and any background noises, providing a comprehensive record of the environment. This system is useful in applications such as surveillance, call monitoring, or assistive technologies where real-time or post-event analysis of audio is necessary. The continuous listening and recording capabilities ensure that no critical audio information is missed, addressing the problem of intermittent or manual recording methods that may fail to capture important details. The system may also include additional components for processing, storing, or transmitting the recorded data, depending on the specific implementation.
16. The system of claim 15 , wherein collecting multi-sensory inputs associated with the voice communication further comprises: obtaining a transcript of a conversation or speech included in the recording by transcribing the conversation or speech using a speech-to-text process; and applying one or more text-based learned models of the at least one learned classifier to the transcript to obtain the one or more classification results, wherein the one or more classification results comprise one or more predicted classification classes and one or more corresponding confidence levels, the overall risk assessment metric is based on the one or more predicted classification classes and the one or more corresponding confidence levels, and the one or more predicted classification classes comprise at least one of the following: a predicted scam type, a predicted conversation signature of a trusted person of the callee, a predicted caller identity of the caller, a predicted call purpose, a predicted social engineering tactic, and a list of suspicious keywords.
This invention relates to a system for analyzing voice communications to assess risk, particularly in detecting scams, fraud, or suspicious activities. The system collects multi-sensory inputs from a recorded voice communication, including audio features, speaker characteristics, and contextual data. It processes these inputs using machine learning models to generate classification results, which include predicted categories such as scam type, caller identity, call purpose, social engineering tactics, and suspicious keywords. The system also transcribes the conversation into text and applies text-based learned models to analyze the transcript, producing predicted classification classes with corresponding confidence levels. These results contribute to an overall risk assessment metric, which quantifies the likelihood of the call being fraudulent or malicious. The system may also compare the conversation against known patterns of trusted individuals to verify authenticity. By integrating multiple data sources and advanced classification techniques, the system enhances the detection of deceptive or harmful communication patterns in real-time or post-call analysis.
17. The system of claim 15 , wherein collecting multi-sensory inputs associated with the voice communication further comprises: applying one or more sound-based learned models of the at least one learned classifier to the recording to obtain the one or more classification results, wherein the one or more classification results comprise one or more predicted classification classes and one or more corresponding confidence levels, the overall risk assessment metric is based on the one or more predicted classification classes and the one or more corresponding confidence levels, and the one or more predicted classification classes comprise at least one of the following: a predicted caller identity of the caller, a predicted emotion of the callee, and a predicted emotion of an adversary.
The system is designed for analyzing voice communications to assess risk by processing multi-sensory inputs, including audio recordings. The system applies sound-based learned models, part of a broader classification framework, to the recorded voice communication. These models generate classification results, which include predicted classification classes and corresponding confidence levels. The predicted classes may identify the caller's identity, the callee's emotional state, or the emotional state of an adversary. The system then uses these results to compute an overall risk assessment metric, which quantifies the likelihood of a security or fraud risk based on the detected patterns. The learned models are trained to recognize specific acoustic and linguistic features that correlate with high-risk scenarios, such as deception, emotional distress, or unauthorized access attempts. By integrating these predictions, the system provides a real-time or post-call risk evaluation to enhance security in communication systems. The approach leverages machine learning to improve accuracy in detecting threats that may not be apparent through traditional voice analysis methods.
18. The system of claim 12 , wherein the at least one sensor device comprises at least one of the following: one or more wearable corporal sensor devices configured to capture the at least one biometric measure of the callee, and one or more other sensor devices deployed in an object of the callee to detect the at least one action carried out by the callee.
This invention relates to a system for monitoring and analyzing biometric and behavioral data of a callee, such as a person receiving a call or communication. The system addresses the problem of determining the callee's availability, attention, or physiological state during interactions, which is critical for applications like emergency response, healthcare monitoring, or call routing optimization. The system includes at least one sensor device configured to capture biometric measures of the callee, such as heart rate, respiration, or movement, and detect actions performed by the callee, such as gestures or object interactions. The sensor devices may be wearable, attached to the callee's body, or embedded in objects the callee interacts with, such as a phone, vehicle, or smart home device. The system processes the sensor data to assess the callee's state, such as stress levels, distraction, or physical activity, and uses this information to determine appropriate responses, such as prioritizing calls, triggering alerts, or adjusting communication settings. The system may also integrate multiple sensor inputs to provide a comprehensive analysis of the callee's condition, improving decision-making in real-time scenarios. This approach enhances situational awareness and enables more effective and context-aware communication systems.
19. The system of claim 18 , wherein the one or more classification results comprise one or more predicted classification classes and one or more corresponding confidence levels, and the overall risk assessment metric is based on the one or more predicted classification classes and the one or more corresponding confidence levels.
This invention relates to a system for assessing risk in data processing, particularly in scenarios where data is classified into categories with associated confidence levels. The system addresses the challenge of accurately evaluating risk when dealing with uncertain or probabilistic classification results, ensuring that risk assessments account for both the predicted class and the confidence in those predictions. The system processes input data through a classification module that generates one or more predicted classification classes along with corresponding confidence levels for each class. These classification results are then used to compute an overall risk assessment metric. The metric is derived by considering both the predicted classes and their associated confidence levels, allowing for a more nuanced and accurate risk evaluation. This approach improves decision-making in applications where classification uncertainty directly impacts risk, such as fraud detection, cybersecurity, or medical diagnostics. The system may also include additional components, such as a data preprocessing module to prepare input data for classification and a risk mitigation module to apply corrective actions based on the computed risk assessment. The integration of confidence levels into the risk assessment ensures that higher uncertainty in classifications leads to more conservative or cautious risk evaluations, reducing the likelihood of erroneous decisions. This method enhances the reliability and robustness of risk assessment systems in real-world applications.
20. A computer program product comprising a computer-readable hardware storage medium having program code embodied therewith, the program code being executable by a computer to implement a method comprising: intercepting a voice communication from a caller to a callee; collecting multi-sensory inputs associated with the voice communication by capturing data via at least one sensor device, applying at least one learned classifier to the captured data, and aggregating one or more classification results from the at least one learned classifier as the multi-sensory inputs, wherein the captured data comprises at least one biometric measure of the callee, and the multi-sensory inputs are indicative of content of the voice communication, at least one action carried out by the callee in response to the content of the voice communication, and a stress level of the callee in response to the content of the voice communication; and determining an overall risk assessment metric for the voice communication based on the multi-sensory inputs and learned signatures, wherein the overall risk assessment metric indicates a likelihood the voice communication is a scam.
This invention relates to a system for detecting fraudulent voice communications, such as scam calls, by analyzing multi-sensory inputs and biometric data from the callee. The system intercepts a voice call between a caller and a callee and collects data from sensor devices, such as cameras, microphones, or biometric sensors, to capture the callee's physical and behavioral responses. A learned classifier processes this data to identify patterns, including the callee's stress level, actions taken in response to the call, and the content of the conversation. The system aggregates these classifications to assess the likelihood that the call is a scam. The risk assessment is based on learned signatures—predefined patterns associated with fraudulent behavior—and generates an overall risk metric indicating the probability of a scam. This approach enhances fraud detection by combining real-time biometric and contextual analysis, improving accuracy over traditional voice-based detection methods. The system is implemented as a computer program product, executing on a hardware storage medium to perform these steps.
Unknown
October 20, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.